IT IS BECOMING F*CKING SCARY: CLAUDE NOW REMEMBERS EVERY PROJECT YOU'VE EVER WORKED ON, AND THE WHOLE MEMORY IS JUST A FOLDER ON YOUR DISK
it's called Basic Memory. 4,200 stars on GitHub, open source
you know that moment you open a new Claude session and burn the first 10 minutes re-explaining your project, your decisions and the bug you fixed last week?
this kills it
Basic Memory is an MCP server that gives Claude a second brain made of plain markdown files
Claude takes notes while you work. every note has 2 parts:
→ observations: [bug], [fix], [decision], [idea]
→ relations: wikilinks like [[api-client]] and [[bug-4471]]
the links stitch every note into a knowledge graph
next session, Claude doesn't start from 0. it calls 3 tools:
> search_notes → finds notes by meaning, not just keywords
> build_context → follows the links and pulls in everything connected
> write_note → saves what it learned today
and it's not a black box. open the same folder in Obsidian or VS Code. edit a note yourself and Claude sees the change
local-first. plain text. 0 lock-in
works with Claude Code, Claude Desktop, Codex, Cursor, VS Code, ChatGPT and anything that speaks MCP
1 line for Claude Code:
claude mcp add basic-memory -- uvx --prerelease=allow basic-memory mcp
(local is free under AGPL. optional paid cloud if you want sync across devices)
give it 1 project and 1 week. then open the folder and read what Claude knows about you
repo below
REA is most dangerous thing on the internet right now...
It lets your AI agent reverse engineer any app, down to the binary
If you use it correctly, you can rebuild any feature you pay for, for FREE
In this article, I show you how: https://t.co/3P1FNXiXzl
THIS IS F*CKING ILLEGAL-LEVEL PRODUCTIVITY: I FIRED MYSELF FROM WATCHING CLAUDE CODE AND HIRED A WHOLE SQUAD OF AI CODERS INSTEAD
it's called Claude Squad. 8,600 stars on GitHub
1 terminal. 6 kinds of coders on payroll: Claude Code, Codex, Gemini, Aider, OpenCode, Amp
the trick is git worktrees
every task gets its own worktree on its own branch. nobody steps on anybody's code
→ Claude fixes the auth bug on branch 1
→ Codex writes billing tests on branch 2
→ Gemini rebuilds the API client on branch 3
→ Aider updates the docs on branch 4
and you run the whole team from 1 tmux window:
> n → hire a new agent
> ↵ → walk over and give it new orders
> tab → flip between live output and the diff
> c → commit and pause
> s → commit and push the branch
nothing ships until the boss reads the diff
there's even an experimental auto-yes mode (-y) that keeps the team working while you're out of office
assign a model per session:
> cs → Claude Code
> cs -p "codex" → Codex
> cs -p "gemini" → Gemini
brew install claude-squad
(needs tmux and gh)
stop being the guy who watches 1 agent type. be the guy who signs off on 4.
repo in the reply
THIS IS F*CKING UNREAL! AI AGENTS ARE REBUILDING A 1996 WINDOWS GAME WITH NOTHING BUT THE .EXE
the game is DX-Ball. no source code. no docs. just the original executable
and REA, the MIT-licensed reverse engineering MCP sitting at 238,000 stars on GitHub
the goal: a clean, maintainable C version of the whole game, recovered 1 function at a time
here's 1 function, start to finish
question: when a brick breaks, which speaker plays the sound?
the agent asked REA 3 things:
→ analyze_function 0x406400 - the decompiler said the function was empty. the raw instructions showed a hidden integer input
→ follow the caller - the brick-hit code turns a tile into a screen position: 20 + 30 × tile_x
→ read_bytes - 2 constants buried in the .exe decode to 1.5625 and 500.0
the whole formula:
pan = (x × 1.5625 − 500.0) × scale
-500 on the left edge. 0 in the middle. +500 on the right
then it had to prove it wasn't guessing:
> 3,205 test cases run against the original x86 code
> every position from 0 to 640, at 5 pan scales
> recompiled with the 1996-era compiler: all 63 bytes identical
the project so far:
> 55 functions rewritten in C
> 45,380 checks against the original
> 33 functions that compile byte-for-byte the same
and it's not just old games. REA covers 23 target types: native binaries through Hopper, Ghidra or IDA, Electron apps, .NET, Android APKs, websites, firmware, network captures
1 command to start:
npx rea-agents setup
then hand your agent a binary you're allowed to inspect and paste this:
"Use REA to recover 1 function from this executable
1. Analyze the function and its callers. Show the instructions behind every claim, not just the pseudocode
2. Read and decode every constant or global it depends on
3. Write a C version with the same inputs and outputs
4. Build a test that runs the original and your version across the full input range. Report every mismatch"
start with 1 small function you can fully verify
the decompiler said empty. the bytes said otherwise.
REA is most dangerous thing on the internet right now...
It lets your AI agent reverse engineer any app, down to the binary
If you use it correctly, you can rebuild any feature you pay for, for FREE
In this article, I show you how: https://t.co/3P1FNXiXzl
YOUR CODING AGENTS KEEP A F*CKING RECEIPT FOR EVERY TOKEN YOU BURN. ALMOST NOBODY HAS EVER OPENED IT
the logs are already on your disk. ccusage reads them and turns them into 1 bill
19,000 stars on GitHub. 1 command. nothing to install:
npx ccusage@latest
and it's not just Claude Code. it reads 18 agents:
→ Claude Code, Codex, OpenCode, Amp, Droid, Codebuff
→ Gemini CLI, Copilot CLI, Grok Build CLI, Qwen, Kimi
→ Goose, Kilo, Hermes Agent, pi-agent, OpenClaw, Antigravity, ZCode
then it slices the bill 5 ways:
> ccusage daily → cost per day, every agent in 1 table
> ccusage session → the exact chat that ate your budget
> ccusage blocks → Claude Code's 5-hour billing windows
> ccusage claude daily --instances → spend per project
> ccusage daily --by-agent --json → raw numbers for your own dashboard
there's even a statusline mode (beta) that shows the meter live inside Claude Code
100% local. nothing leaves your machine
run 1 daily report before you upgrade your plan
you're not paying for the model. you're paying for the chats you forgot about.
bro your $6.80 cow is f*cking running on its tits
I paid $50. mine has 4 actual legs. 4 f*cking knees. a cowbell.
it jumps a fence, then clears a whole hay wall in bullet time while the camera spins around it like the Matrix
lands on its hooves. dust everywhere. moos at the end.
7x the price. 0 udders used as legs.
What the f*ck is this drug trip
I told Opus 5.5 to make a cow that runs. This is what it gave me for $6.80.
It runs on its f*cking udder. Four teats as legs. Then it jumps a fence in slow motion and lands like a pancake.
"BLOCKED BY KORNET!"
A play that captures what NBA Defense Week is all about... Luke Kornet's CLUTCH SWAT in Game 7 of the Western Conference Finals 🖐️😳
Which was your favorite defensive play of the season?
🎆2026 https://t.co/5xo7LSIrN4 New Year Airdrop🎉
Snag your slice of 10,000 $ANB rewards by finishing a few quick tasks:
🔶Connect your X account and follow @ant_fun_trade
🔶Retweet our post
🔶Make your first trade on Web (0 fees!) or in the App
🔶Invite friends and earn extra bonuses
Event ends in 15 days, check the details on the event page.
Join now: https://t.co/YgHTiUmKnk
Fully Local Wallet Control + Multi-Chain All-in-One + Zero Platform Fee = Your Future Trading!
2000 USDT GIVEAWAY 💥
The market is green - and Yeti is giving back again!
🎁 Prize Pool: 2000 USDT
💸 20 winners = 100 USDT each
How to enter:
1. Follow @yeti_dyor
2. Like this post
3. Retweet this post
4. Tag 3 trader friends in the comments
📦 Equal chances for all - winners randomly picked on July 24.
Every winner will receive $100.
Let’s celebrate the green together!
Good luck, legends! 😎